ResearchBenchmarksFactual mistake detection on DocENT (PoSh)Factual mistake detection on DocENT (PoSh)Follow64.8AccuracySoft-MSD34.01642.0085057.992May 7, 2026Evaluation resultsMethodMethodLinksAccuracySpearman Rho (ρ)Kendall Tau (τ)Soft-MSD2026.0564.80.2590.195Soft-MSD2026.0564.40.2190.165CLIPScore2026.0553.50.1810.136CLIPScore2026.0545.30.1450.108FLEUR2026.0541.20.040.031FLEUR2026.0535.20.0530.04